CovidXrayNet

CovidXrayNet performs automated detection of COVID-19 from chest X-ray (CXR) images by optimizing data augmentation and convolutional neural network (CNN) hyperparameters to improve diagnostic accuracy.

Key Features:

• Optimized Data Augmentation: A refined data augmentation pipeline applied to CXR images to improve CNN model robustness and generalization.
• CNN Hyperparameter Optimization: Fine-tuning of CNN hyperparameters to maximize validation accuracy, yielding reported improvements of 11.93% for VGG-19 and 4.97% for ResNet-50.
• EfficientNet-B0 Architecture: Implements the EfficientNet-B0 architecture as a primary model benefiting from augmentation and hyperparameter optimizations.
• Three-class Classification: Performs classification into COVID-19, normal, and pneumonia classes from CXR images.
• Performance Metrics: Reported accuracy of 95.82% on the COVIDx dataset after 30 training epochs.

Scientific Applications:

• Medical imaging diagnostics: AI-driven screening of COVID-19 from chest X-rays to distinguish COVID-19, normal, and pneumonia cases for diagnostic support.

Methodology:

Apply optimized data augmentation and CNN hyperparameter tuning, evaluate models including EfficientNet-B0, VGG-19, and ResNet-50 on COVIDcxr (960 CXR images) and COVIDx (15,496 CXR images) in a three-class classification task (COVID-19, normal, pneumonia); report 95.82% accuracy on COVIDx after 30 epochs and improvements of 11.93% (VGG-19) and 4.97% (ResNet-50).